README.md

Human-pose-estimation

A quick tutorial on multi-pose estimation with OpenCV, Tensorflow and MoveNet lightning.

**Also available on Kaggle

Motivation

I found multipose estimation Notebooks and codes not so explicit or even understandable for pure beginners. Moreover, most of the available tutorials focus on single-pose estimation, with only one instance (human). As a result, the idea of writing my own tutorial naturally came to me. After some research and a bit of styling, code cleaning, presentation...I finally made it public.

Model info

• Model type : MoveNet
• Pose estimation method : multipose, bottom-up
• Keypoint count : 17

Results (example)

How to use

Command line runner

• Test script
• Bat file for a sample test : requires the model path to be ../Model/TFLite/lite-model_movenet_multipose_lightning_tflite_float16_1.tflite. You can download it directly from TFHub

Args :

Notebook

• Kaggle

Requires Tensorflow Hub and a Kaggle environment. However, feel free to adapt to Notebook to your local setup

Required packages

• Deep Learning and calculations : Tensorflow 2.x, NumPy, Tensorflow Docs
• Computer graphics/vision : OpenCV
• Display : IPython, Matplotlib
• Image/video writer : image io

Acknowledgements

• Tensorflow official tutorial